Improving semantic similarity of words by retrofitting word vectors in sense level

This paper presents an approach for retrofitting pre-trained word representations into sense level representations to improve semantic distinction of words. We use semantic relations as positive and negative examples to refine the results of a pre-trained model instead of integrating them into the objective functions used during training. We experimentally evaluate our approach on two word similarity tasks by retrofitting six datasets generated from three widely used techniques for word representation using two different strategies. Our approach significantly and consistently outperforms three state-of-the-art retrofitting approaches.